Choosing the best competition models using maximum likelihood tests

Chih-ching Yu, Purdue University

Abstract

Statistical testing of market structure-economic performance relationships began with an article by Joe Bain published in 1951. Estimating market structure-performance relationships usually requires some conduct assumptions prior to empirical testing. Consequently, one notable problem in empirical industrial organization research is the sensitivity of performance conclusions to different conduct assumptions. The choice of appropriate conduct assumptions has implications for public policy and enterprise strategy. Quang Wong's (1989) maximum likelihood test based on the Kullback-Leiber Information Criterion is a promising and highly general test for model selection that can be applied to nested, non-nested and overlapping models. In this dissertation, three competition models were estimated to attempt to assess the feasibility of Wong's (1989) maximum likelihood test in finding the best model in the U.S. margarine industry from 1989 to 1996. The first order conditions were derived for three models: pure competition, noncooperative oligopoly, and cooperative oligopoly. The initial application to alternative models that included four firms proved to be unmanageable due to excessive computational burden. However, the maximum likelihood test for the two oligopoly models containing two firms was successful and showed that the best hypothesis given the market structure of the margarine industry in the U.S. was cooperative oligopoly. The hypothesis of noncooperative oligopoly behavior for the U.S. margarine industry was rejected. The results of this research imply that Vuong's maximum likelihood test for model selection can be used to assist in reducing the burden of proof in public antitrust enforcement. In a practical setting, once the necessary data are collected from the investigated business entities and are submitted to either a public antitrust authority or private plaintiffs, the procedure of choosing the most appropriate market structure for the inspected industry could be completed in approximately ten days, assuming sufficient computational capacity. As a data-driven method for distinguishing legal from illegal behavior, this test has the potential for reducing the scope of disagreement between parties. In that case, it may not only save financial resources in antitrust enforcement but also could save time in assessing competitiveness and the size of fines or injuries. Furthermore, Vuong's maximum likelihood test can be used as one of the business managers' market analysis tools for existing firms as well as potential entrants.

Degree

Ph.D.

Advisors

Connor, Purdue University.

Subject Area

Agricultural economics|Marketing

COinS